Epigenetic clocks: what can Horvath, GrimAge and DunedinPACE do?

What are epigenetic clocks?

Epigenetic clocks are algorithms that use patterns in DNA methylation to estimate age, risk, or rate of aging. They are based on the idea that specific methylation sites in the genome change with age and with biological stress over time. PMID 24138928 PMID 36131109

However, this does not mean that a clock is a direct window into your entire physiology. A clock is a statistical model that is trained for a specific purpose. Therefore, it is crucial to know whether the model is built to predict chronological age, healthspan-close outcomes or tempo in aging. PMID 24138928 PMID 36131109

Horvath, GrimAge and DunedinPACE do not answer the same question

Many readers think that the three names are just different brands for the same type of biological age test. It is wrong. The difference lies in what the models were trained to capture, and therefore also in how their output should be interpreted. PMID 24138928 PMID 29676998

The simplest way to look at it is that Horvath asked how old the profile looks, GrimAge asked more directly about healthspan and mortality near-risk, and DunedinPACE asked how fast the aging process seems to run. PMID 24138928 PMID 29676998

What the Horvath clock actually did

The 2013 Horvath clock became pivotal because it showed that a relatively small set of methylation sites could be used to estimate age across many human tissues and cell types. This made it an important reference point for the entire field and one of the most cited epigenetic clocks. PMID 24138928

The important thing today, however, is to understand that Horvath was primarily known as a pan-tissue age model. This makes it useful as a conceptual basis and as a broad comparison framework, but not automatically as the best tool for direct decisions about risk, intervention or mortality. PMID 24138928

Why GrimAge got so much attention

GrimAge became known because the model was developed to be closer to healthspan and mortality than the early age clocks. It combines DNA methylation-based surrogates for, among other things, plasma proteins and smoking pack-years to get closer to outcomes that mean more in the real world than a pure age guess. PMID 29676998

This is also why GrimAge is often referred to as more 'clinically interesting' than Horvath. But that doesn't mean GrimAge is a personal divination tool. It's still a statistical model, and it still needs to be held up against classic markers like blood pressure, glucose, fitness, body composition and symptoms. PMID 29676998

Why DunedinPACE is a different kind of marker

DunedinPACE differs by being developed as a pace of aging model. Instead of translating your methylation profile into a biological age number in years, it tries to estimate how fast the aging process seems to be going right now. PMID 30669119

This is an important difference because many readers are looking for a single biological age figure, while in practice DunedinPACE is more interesting for longitudinal follow-up. This is precisely why it makes sense to see DunedinPACE as a supplement to Horvath and GrimAge logic, not just another version of the same dashboard. PMID 30669119

When does a methylation test actually make sense?

For most people, an advanced methylation test makes the most sense when there is already a fairly serious program around sleep, exercise, blood tests and follow-up. Without that framework, the test often becomes mostly an expensive fascination with little action value. PMID 24138928 PMID 29676998 PMID 30669119

If you are not yet working in a structured way with blood pressure, glucose, waist circumference, fitness or strength, it is often better to start there. The methylation test becomes more useful when it is used to qualify an existing process and not as a substitute for the more actionable layers. PMID 24138928 PMID 29676998 PMID 30669119

The biggest mistakes in interpretation

The most common mistake is to call everything biological age and stop there. When age, risk and pace are mixed together, it becomes difficult to understand what the test result actually means. This increases the risk of both hype and unnecessary concern. PMID 36131109

The second biggest mistake is jumping straight to the most advanced test without mastering the basic markers. An elegant methylation report cannot compensate for a lack of overview of blood pressure, glucose, fitness, sleep or body composition. PMID 36131109

The three generations of DNA methylation clocks: Chronology, mortality, and pacing

The maturation of epigenetic clocks over the past fifteen years marks one of the most significant revolutions in objective biogerontology. The field is methodologically classified into three distinct algorithmic generations: PMID 24138928 PMID 29676998 PMID 30669119 PMID 35029144

1. **First-Generation Clocks (2013)**: Spearheaded by Steve Horvath (353 multi-tissue CpG sites) and Gregory Hannum (71 whole-blood CpG loci). These models were trained directly against **chronological calendar age**. While confirming that aging imprints a precise mathematical signature upon the methylome, they harbored a fatal clinical weakness: They penalized healthy older adults simply for living longer and correlated poorly with prospective morbidity and all-cause mortality. PMID 24138928 PMID 29676998 PMID 30669119 PMID 35029144

2. **Second-Generation Clocks (2018–2019)**: Re-engineered around phenotypic aging and clinical survivorship. Morgan Levine's **DNAm PhenoAge** (513 CpGs) was calibrated against a composite physiological mortality score derived from 9 routine clinical blood analytes. Steve Horvath and Ake Lu's **DNAm GrimAge** (1,030 CpGs) set a new gold standard by training methylation surrogate markers against 7 circulating plasma proteins (including cystatin C, PAI-1, and GDF15) alongside historical pack-years of tobacco exposure. GrimAge remains the most robust molecular predictor of cardiovascular events, cancer onset, and remaining lifespan. PMID 24138928 PMID 29676998 PMID 30669119 PMID 35029144

3. **Third-Generation Clocks (2022–2026)**: Exemplified by **DunedinPACE** and Harvard's **OMICmAge**. Rather than computing static biological years, third-generation algorithms capture the **dynamic velocity** of multi-system deterioration per calendar year derived from longitudinal cohorts and deep circulating proteomics. PMID 24138928 PMID 29676998 PMID 30669119 PMID 35029144

The laboratory assay pipeline: Bisulfite conversion, EPIC arrays, and cellular deconvolution

Interpreting an epigenetic report requires understanding the wet-lab molecular workflow: PMID 24138928 PMID 36131109

High-molecular-weight genomic DNA is extracted from biological specimens—with peripheral blood mononuclear cells (PBMC) or dried blood spot (DBS) capillary blood representing the clinical reference standard. DNA undergoes **sodium bisulfite deamination**, a chemical reaction that selectively converts unmethylated cytosine residues into uracil, while leaving 5-methylcytosines (methylated CpGs) chemically intact. PMID 24138928 PMID 36131109

The bisulfite-treated DNA is subsequently hybridized onto high-density microarrays, typically the **Illumina Infinium MethylationEPIC v2 BeadChip**, which measures fluorescent intensity ratios (beta values from 0.0 to 1.0) across more than 935,000 specific genomic loci. PMID 24138928 PMID 36131109

Crucially, the raw array data is processed via bioinformatic **cellular deconvolution** (such as the Houseman reference-based algorithm). Because peripheral blood contains granulocytes, monocytes, CD4+ and CD8+ T lymphocytes, B cells, and natural killer cells—each carrying distinct epigenetic identities—the algorithm mathematically separates true intracellular epigenetic drift from confounding shifts in leukocyte cell-type proportions (such as acute infectious lymphopenia). PMID 24138928 PMID 36131109

Mitigating batch variance: Principal Component (PC) architecture and longitudinal tracking

Early iterations of consumer epigenetic testing suffered from noticeable laboratory assay variability: Aliquots from an identical blood draw processed across different plate batches or weeks could diverge by 3 to 4 biological years due to minute temperature, reagent, or chip-lot fluctuations. PMID 30669119 PMID 36131109

This reproducibility bottleneck was solved by Higgins-Chen et al. at Yale University in 2022 via the formulation of **Principal Component (PC) clocks** (such as PC-GrimAge, PC-PhenoAge, and PC-Horvath). By projecting thousands of correlated CpGs into orthogonal principal components, technical assay noise is filtered out, compressing test-retest margins below 12 months. PMID 30669119 PMID 36131109

For rigorous serial longevity tracking, optimal protocol entails testing under steady-state baseline health (avoiding testing during acute febrile infections or antibiotic regimens), drawing blood under morning fasting conditions, and retesting at 12- to 18-month intervals utilizing an identical assay platform. PMID 30669119 PMID 36131109